Fix headless CLI startup on macOS

[shanemesa]

Summary

This fixes a macOS CLI startup regression where pluginval could abort before handling command-line options. On macOS 26.4, both the current v1.0.4 release and current develop aborted on direct CLI startup, including pluginval --version, which also prevented headless validation in CI-like usage.

Problem

• Direct CLI launch aborted before normal command handling:
  • pluginval --version
  • pluginval --strictness-level 5 --skip-gui-tests /path/to/plugin.vst3
• The abort happened during app/bootstrap behavior that is unnecessary for CLI mode.
• In local testing, the crash path involved macOS application registration during startup and also reproduced when the CLI path attempted to change dock/presentation state.

Approach

• Route CLI mode before full JUCEApplication startup.
• Keep the GUI app path unchanged.
• Run CLI mode with a minimal headless JUCE/message-loop setup.
• Avoid dock/presentation changes in CLI mode, because they force application registration on recent macOS builds and can break headless execution.
• Add small tests for raw argv CLI detection so the custom entry path stays covered.

Scope

• Source/Main.cpp
• Source/CommandLine.h
• Source/CommandLine.cpp
• Source/CommandLineTests.cpp

Notes

• This PR is intentionally narrow. It fixes CLI startup/routing and does not change validation logic.
• GUI behavior is left on the existing JUCEApplication path.
• AU discovery failures seen locally are separate from this startup fix.

Repro

Before this patch, on macOS 26.4:

pluginval --version

[drowaudio]

Are you sure all of this is required? It seems like a lot of changes just to fix version reporting...
I can understand not calling hideDock, but why can't we have a normal juce application running headless?

[shanemesa]

Are you sure all of this is required? It seems like a lot of changes just to fix version reporting... I can understand not calling hideDock, but why can't we have a normal juce application running headless?

Thanks, that’s fair.

--version was just the smallest repro. The real issue on my machine was that CLI startup in general was aborting on macOS 26.4, including --validate, so it was breaking headless validation, not just version reporting.

From the crash report, the abort was happening during JUCEApplicationBase::main / NSApplication startup, before initialise() ran. That’s why I didn’t stop at the dock call.

That said, I agree the patch may be broader than ideal. If you’d prefer, I can rework it into a smaller version that:

• keeps the normal GUI app path unchanged
• removes the dock/presentation change in CLI mode
• keeps --version, --help, --run-tests, and --validate working
• still exits cleanly with the right return code

I’d only want to do that if it still fixes real --validate runs on my machine, not just --version.